jarvis-console

Jarvis Web Console

Build and manage the Jarvis command center with agentic task execution, MCP Console, AI Assist, local model inference, and agent mesh.

Environment Detection

This skill requires Docker and local shell access. Before starting:

• Claude Code (CLI): Full access — can rebuild, deploy, verify containers → proceed with all steps
• Claude Cowork (web/desktop): No Docker, no local shell → can only review/edit config files (mcp_config.json, version-history.json, main.py). Tell Chadron: "Console rebuild requires a Code session. I can update configs and version files now — rebuild in your next Code session."

Context

• Workspace root: resolved from CLAUDE.md canonical root. In Cowork sessions, use mounted workspace.
• Console project location: Jarvis/console/
• Management script: jarvis-console (commands: start/stop/restart/rebuild/status/logs/open/api/nuke)
• MCP config source of truth: Jarvis/console/backend/config/mcp_config.json
• Version history: Jarvis/console/backend/config/version-history.json

Current Version: 2.49.0

Component	Runtime	Port	Notes
Backend	Docker (console-backend-1)	8000	FastAPI — reads brain files via mounted volume
Frontend	Docker (console-frontend-1, nginx)	3000	React + Vite + Tailwind dark theme
WebSocket	Docker backend	8000/ws	Real-time task streaming

CRITICAL: Backend runs in Docker, NOT on host. nginx proxy_pass uses http://host.docker.internal:8000. load_dotenv(override=False) so container env vars win. Health probes at both /health/live and /api/health/live.

Key Features

• MCP Console: unified single-input (tool name → structured, NL → AI Assist via Claude Sonnet)
• Local Model Inference: routes commodity tasks to Ollama via delegate.py; endpoint POST /api/inference
• Agent Mesh: multi-node mesh health + routing (mesh.py, agents.py, agent_orchestration.py)
• Semantic Memory: Qdrant-backed vector memory for cross-session context (qdrant_memory.py)
• Scheduler: cron-based task scheduling with scheduler_service.py
• GitHub Live Data: /api/github/releases + /api/github/milestones pull live (no local JSON cache)
• Token Analytics: per-model cost tracking via model_usage_tracker.py
• Wave Tracking: milestones from GitHub, scored planned waves from config/wave-priorities.json

Backend Routers (30+)

routers/
  status.py, projects.py, recommendations.py, needs.py, history.py
  growth.py, tasks.py, graph.py, audit.py, monitoring.py
  mcp.py, memory.py, inference.py, ollama.py, mesh.py
  agents.py, agent_orchestration.py, benchmarks.py, github.py
  ai.py, local_ai.py, docs.py, data.py, catalog.py
  agile.py, governance.py, context.py, improvements.py
  portal.py, scheduler.py, search.py, system.py
  token_analytics.py, nlq.py, workbook.py, plugin.py

Key Services

services/
  brain_reader.py       # Brain file parser + MCP health check
  mcp_executor.py       # MCP tool introspection + execution
  mcp_router_client.py  # jarvis-router HTTP client
  nl_interpreter.py     # Claude Sonnet NL → tool mapping
  agent_memory.py       # Semantic memory operations
  qdrant_memory.py      # Vector store integration
  semantic_memory.py    # Memory retrieval engine
  scheduler_service.py  # Cron-based task scheduling
  model_usage_tracker.py # Token + cost analytics
  ai_service.py          # AI service abstraction

Console Version Bump Protocol

Update ALL FOUR locations together:

1. Jarvis/console/backend/main.py → version="X.Y.Z"
2. Jarvis/status.json → web_console.version
3. Jarvis/console/backend/config/version-history.json → insert new entry at array top
4. Jarvis/console/backend/config/mcp_config.json → if tool counts changed

KPICards and Sidebar read from /api/status — never hardcode versions in JSX.

Docker Rebuild (required after any backend or config change)

cd /Users/chadronbryant/infrastructure
docker compose -f docker-compose.headless.yml --env-file .env build --no-cache backend frontend
docker compose -f docker-compose.headless.yml --env-file .env up -d --force-recreate backend frontend

Verify: curl http://localhost:8000/api/health → version must match.

Console Stability Guardrails

1. nginx proxy_pass: always http://host.docker.internal:8000 (not Docker sibling DNS)
2. load_dotenv(override=False) in main.py — container env vars must win
3. Health probes: keep both /health/live AND /api/health/live (compatibility aliases)
4. Frontend instance selection: persisted in localStorage key jarvis_active_instance_id
5. Local inference routing: prefer_local: true or source = codex-cli / antigravity / delegate-cli

Verifier: bash scripts/jarvis/verify_console_guardrails.sh